[R] Different missing links on Windows in 'check' vs. 'install'

[Duncan Murdoch]

On Tue, 9 Mar 2004 12:38:33 +0000 (GMT), Prof Brian Ripley
<ripley at stats.ox.ac.uk> wrote :

>On Tue, 9 Mar 2004, Duncan Murdoch wrote:
>> Here's a proposal:
>> 
>> All of check and build and install should default to the same library
>> location.  Check and build aren't meant to be permanent installs, so
>> if the package already exists there, it'll have to be temporarily
>> moved out of the way.
>
>You may not own the main library and so not have permission to 
>install/check/build there.  If you do, you can use check on an installed 
>copy of the package.  As for build, this is one of the reasons why
>Rcmd INSTALL --build was needed, as that installs in the standard place 
>and then wraps up the installed copy.

This isn't a revolutionary proposal, I'm mainly suggesting that we
regularize it a bit.  If I want to build a binary package, I should be
able to expect to use "Rcmd build --binary".  If "Rcmd INSTALL
--build" does a better job, then shouldn't "Rcmd build --binary" be
modified to do the same?

The only new part of my suggestion is to move existing packages out of
the way so that the new one can be installed temporarily.  As you say,
this won't always be possible, but if "Rcmd INSTALL --build" could
work, then it should be possible to do a temporary version of the same
thing.

So here's a refinement of my suggestion:

By default, Rcmd build and Rcmd check should attempt to do their work
in the default library install location.  If that fails, then they
could fail with a message telling you to use a  --library= option, or
they could automatically try working in a temporary directory, with a
warning that this may produce suboptimal results.

In either case, if a package with the given name is already installed,
they should attempt to rename it to a temporary name to get it out of
the way.  If this fails, then proceed as above.

Duncan Murdoch